aroonwithaview

Like seemingly every other person on the planet today, I grapple daily with the implications of rapid advances in AI. I was too young during the internet age to know how people felt then, but my conjecture is that this feels like a much more profound shift. In more ways than one, it has made us reflect on what it means to be human, and the answers aren’t always convincing or comforting.

I do find the doomers entirely unconvincing, however. Claims that humans will be rendered obsolete feel more like science fiction than sound reason. But I also think that thought exercises like the “country of geniuses” that Amodei invokes in his Adoloscence of Technology essay are too ambitious. And in a sense, I think both sides are wrong for very similar reasons.

In both views of the world, a powerful system of AI emerges, largely unconstrained by the physical world. But resource constraints in the physical world are very real – ask Nvidia or TSMC, who have not been able to provide sufficient compute for even the somewhat smart AI of today. And these resource constraints become even more acute when we get to truly useful “embodied” AI – AI that can actually interact with the real world.

In fact, I find it really useful to think of AI today as a magically intelligent newborn. Each instance of it has insufficient memory and no way to interact with the real world. And I find it hard to believe that these characteristics will improve in any significant way in the near future, unless we throw an unsustainable amount of resources at it or embody it in an expensive robot. From this, I believe that humans will continue to be kings of context and the physical world, at least in the short run. Having lived, and being able to remember having lived, for decades gives us the edge – and the people who have lived more and remember more (and contextualize their memory better) will be the winners in this new paradigm. An infant AI needs its granfatherly human.